Abstract:
The article considers the possibility of using weekly composite images from the Meteor-M No. 2 satellite to classify arable lands in Khabarovsk Krai. For four vegetation classes (soybeans, grain crops, perennial grasses, and fallow land), average Normalized Difference Vegetation Index (NDVI) seasonal variation series were constructed for municipal districts in the south of Khabarovsk Krai in 2024 and the main characteristics — the NDVI maximum values and the day of the maximum — were calculated. Statistically significant differences in the indicators for the average NDVI time series for different vegetation classes were revealed ($p< 0.0001$). Using validated data from Khabarovsk KRAI, a classification of arable lands in the Bikinsky, Vyazemsky, and Lazovsky Districts was conducted using machine learning (the Random Forest algorithm). The average accuracy of the method based on the results of three-fold cross-validation was equal to $87.6\%$. For different vegetation classes, the F1 metric value ranged from $0.61$ to $0.93$. Arable land maps were created for the southern regions of Khabarovsk Krai. It was found that fallow land accounts for over $30\%$ of the region's total arable land area, while soybean crops accounted for $48\%$ in 2024. The mapping results were entered into the developed geographic information system.
Fulltext:
PDF file (889 kB)